Reimaging Medicine
Revolutionize cancer immunotherapy by data sciences for precision medicine.
Immunotherapy is one of the main treatments for patients with cancer. While promising benefit is observed in the clinic, less than half of the patients respond across multiple histologies. Here, I ask a simple question: why some patients respond and some do not? Trained as a bioinformatician and data scientist, I answer these questions by establishing a hybrid data science model, marrying large-scale omics/imaging databases with clinically annotated, carefully curated, locally established biobanking and trial cohorts. I use this model to discover tumor resistance pathways that drive immune exclusion and lack of response to cancer immunotherapy. Such knowledge, especially captured at the per-patient level, provides solutions tailored towards each patient’s unique genetic background. In collaboration with my long-term clinician colleagues and physician-scientist friends, many of these findings have been translated into new clinical trials to benefit patients.
Deep learning-assisted computer vision for new biomarker discovery.
Given the rapid evolution of computer vision and scalable cloud computing, extracting signals from thousands of pathology images, both spatially and quantitatively, has become a reality. Most importantly, associating such signals with clinical outcome allows the discovery of new biomarkers for disease or response prediction. Here, I use computer vision to interrogate the interactions between malignant / immune / stroma cells from pathology images to discover new therapeutic targets and patient populations that respond to such targets in cancer immunotherapy. Working together with my brilliant students and trainees, we build new models to improve the signal processing of digital pathology data, with the ultimate goal to perform real-time diagnosis and prediction in the clinic.
Integrate multi-modal, multi-omics data for improved prediction of clinical outcome.
Biology is a complicated system. No single dimension may explain all the variables that shape a biological system. Cancer, in particular, evolves over many years taking advantage of complex host and environmental systems to maintain its malignancy. A key solution to resolve such a puzzle is to understand cancer as a multi-dimensional space, similar to that of a universe, where multiple stars (cells) compose the entire system. To pursue this, I build multivariable predictive models to integrate signals from tumor, blood, stool samples of patients using graph-based machine learning technologies.
The long-standing and rapidly evolving role of the microbiome in human diseases.
Viruses, bacteria, fungi, other small microorganisms have co-existed with us humans since day one. They shape our planet, our environment, and also the immune system in our bodies. Historically, immunologists and medical doctors have used microbe-based interventions to treat human diseases, such as injection of bacteria or oncolytic viruses into tumors to boost immune response, thereby eliminate malignant cells. Rapid development in high-throughput sequencing technologies and computing power have enabled new landmark discoveries in host-microbe interactions and their modulatory role in response to treatment. Here, I leverage sequencing and LC-MS/MS techniques to nominate specific microbiota species and microbiota-derived metabolite compounds that should be prioritized for new therapeutic interventions for patients.
Reproducible, scalable, and modularized bioinformatics workflow for cloud computing.
We are living in an era of data explosion. On a daily basis, we generate a lot of data, socially, economically, and personally. In my work, I process millions of data points for every project, encompassing genomics, clinical, and imaging fields. This would be an impossible job without high computing power and storage. My team develops reproducible, scalable, and modularized bioinformatics workflows built upon Docker and Nextflow that have been robustly benchmarked on the AWS EC2 platform. We collaborate with fantastic teams from UPMC and AWS, to use those workflows for scientific discovery with a strong focus on cancer immunotherapy.